Vercel sells Pro teams $80 of GLM 5.3 and Kimi K3 credit for $30 a month

The headline ratio is 2.67 credit dollars per dollar paid ^[9]. Put differently, the $30 price is 62.5% below the face value of the credit ^[10]. Whether that is a discount on tokens depends on how credit draws down, and the changelog does not say. If GLM 5.3 and Kimi K3 are billed against it at their list rates, the ratio holds. If the draw-down rate is higher than list, the real ratio is lower.

Take the list-rate case. The subscription renews monthly unless canceled, and unused credit does not roll over ^[2], so the $30 is charged whatever the team spends. A team that runs $30 of list-priced tokens through it breaks even, and $30 is 37.5% of the credit ^[11]. Above that the team gains, up to $50 of face value at full use ^[12]. For the ratio to transfer to a given workload, the team's spend on these two models has to land between $30 and $80 a month at list rates, arrive evenly enough that the pool lasts the month, and be work that GLM 5.3 or Kimi K3 can do.

The data terms are the stronger half of the offer. Deletion after each request is promised by two parties, AI Gateway and the provider, and prompts are not used to train models ^[8]. We think that matters most for coding agents, which send source code with every request.

The stop at zero is the right tradeoff for a team that wants a fixed bill. The context is that the bill, not uptime, is the constraint. Teams that run agents against a deadline may see it differently, because a refused request late in a billing period stays refused until the next one ^[3]. runtimewire's summary is that the monthly cap and non-rolling credits set a fixed allowance but limit how much a team can rely on the bundled credits ^[5].